How about adding an option to require an alternate allele to appear in at least a certain fraction of forward vs. reverse reads?
I added this after your comment. See the SAB (strand bias alternate) and SRB (strand bias reference) tags in the VCF output. You can filter on this using vcffilter: https://github.com/ekg/vcflib/blob/master/vcffilter.cpp
I'm not sure this is necessary. Now SAP and SRP encode the same values. Furthermore, in default behavior, freebayes incorporates strand balance expectations into its Bayesian genotyping algorithm, which in practice manages much of this problem without reducing sensitivity through a hard input filter.